mirror of
https://github.com/comfyanonymous/ComfyUI.git
synced 2025-06-08 12:35:30 +08:00
dea1c7474a
* Add Ideogram generate node. * Add staging api. * COMFY_API_NODE_NAME node property * switch to boolean flag and use original node name for id * add optional to type * Add API_NODE and common error for missing auth token (#5) * Add Minimax Video Generation + Async Task queue polling example (#6) * [Minimax] Show video preview and embed workflow in ouput (#7) * [API Nodes] Send empty request body instead of empty dictionary. (#8) * Fixed: removed function from rebase. * Add pydantic. * Remove uv.lock * Remove polling operations. * Update stubs workflow. * Remove polling comments. * Update stubs. * Use pydantic v2. * Use pydantic v2. * Add basic OpenAITextToImage node * Add. * convert image to tensor. * Improve types. * Ruff. * Push tests. * Handle multi-form data. - Don't set content-type for multi-part/form - Use data field instead of JSON * Change to api.comfy.org * Handle error code 409. * separate out nodes per openai model * Update error message. * fix wrong output type * re-categorize nodes, remove ideogram (for now) * oops, fix mappings * fix ruff * Update frontend to 1.17.9 * embargo lift rename nodes * remove unused autogenerated model code * fix API type error and add b64 support for 4o * fix ruff * oops forgot mask scaling code * Remove unused types. --------- Co-authored-by: bymyself <cbyrne@comfy.org> Co-authored-by: Yoland Y <4950057+yoland68@users.noreply.github.com> Co-authored-by: thot-experiment <thot@thiic.cc>
426 lines
14 KiB
Python
426 lines
14 KiB
Python
import io
|
||
from inspect import cleandoc
|
||
|
||
from comfy.utils import common_upscale
|
||
from comfy.comfy_types.node_typing import IO, ComfyNodeABC, InputTypeDict
|
||
from comfy_api_nodes.apis import (
|
||
OpenAIImageGenerationRequest,
|
||
OpenAIImageEditRequest,
|
||
OpenAIImageGenerationResponse
|
||
)
|
||
from comfy_api_nodes.apis.client import ApiEndpoint, HttpMethod, SynchronousOperation
|
||
|
||
import numpy as np
|
||
from PIL import Image
|
||
import requests
|
||
import torch
|
||
import math
|
||
import base64
|
||
|
||
def downscale_input(image):
|
||
samples = image.movedim(-1,1)
|
||
#downscaling input images to roughly the same size as the outputs
|
||
total = int(1536 * 1024)
|
||
scale_by = math.sqrt(total / (samples.shape[3] * samples.shape[2]))
|
||
if scale_by >= 1:
|
||
return image
|
||
width = round(samples.shape[3] * scale_by)
|
||
height = round(samples.shape[2] * scale_by)
|
||
|
||
s = common_upscale(samples, width, height, "lanczos", "disabled")
|
||
s = s.movedim(1,-1)
|
||
return s
|
||
|
||
def validate_and_cast_response (response):
|
||
# validate raw JSON response
|
||
data = response.data
|
||
if not data or len(data) == 0:
|
||
raise Exception("No images returned from API endpoint")
|
||
|
||
# Get base64 image data
|
||
image_url = data[0].url
|
||
b64_data = data[0].b64_json
|
||
if not image_url and not b64_data:
|
||
raise Exception("No image was generated in the response")
|
||
|
||
if b64_data:
|
||
img_data = base64.b64decode(b64_data)
|
||
img = Image.open(io.BytesIO(img_data))
|
||
|
||
elif image_url:
|
||
img_response = requests.get(image_url)
|
||
if img_response.status_code != 200:
|
||
raise Exception("Failed to download the image")
|
||
img = Image.open(io.BytesIO(img_response.content))
|
||
|
||
img = img.convert("RGB") # Ensure RGB format
|
||
|
||
# Convert to numpy array, normalize to float32 between 0 and 1
|
||
img_array = np.array(img).astype(np.float32) / 255.0
|
||
|
||
# Convert to torch tensor and add batch dimension
|
||
return torch.from_numpy(img_array)[None,]
|
||
|
||
class OpenAIDalle2(ComfyNodeABC):
|
||
"""
|
||
Generates images synchronously via OpenAI's DALL·E 2 endpoint.
|
||
|
||
Uses the proxy at /proxy/openai/images/generations. Returned URLs are short‑lived,
|
||
so download or cache results if you need to keep them.
|
||
"""
|
||
def __init__(self):
|
||
pass
|
||
|
||
@classmethod
|
||
def INPUT_TYPES(cls) -> InputTypeDict:
|
||
return {
|
||
"required": {
|
||
"prompt": (IO.STRING, {
|
||
"multiline": True,
|
||
"default": "",
|
||
"tooltip": "Text prompt for DALL·E",
|
||
}),
|
||
},
|
||
"optional": {
|
||
"seed": (IO.INT, {
|
||
"default": 0,
|
||
"min": 0,
|
||
"max": 2**31-1,
|
||
"step": 1,
|
||
"display": "number",
|
||
"tooltip": "not implemented yet in backend",
|
||
}),
|
||
"size": (IO.COMBO, {
|
||
"options": ["256x256", "512x512", "1024x1024"],
|
||
"default": "1024x1024",
|
||
"tooltip": "Image size",
|
||
}),
|
||
"n": (IO.INT, {
|
||
"default": 1,
|
||
"min": 1,
|
||
"max": 8,
|
||
"step": 1,
|
||
"display": "number",
|
||
"tooltip": "How many images to generate",
|
||
}),
|
||
"image": (IO.IMAGE, {
|
||
"default": None,
|
||
"tooltip": "Optional reference image for image editing.",
|
||
}),
|
||
"mask": (IO.MASK, {
|
||
"default": None,
|
||
"tooltip": "Optional mask for inpainting (white areas will be replaced)",
|
||
}),
|
||
},
|
||
"hidden": {
|
||
"auth_token": "AUTH_TOKEN_COMFY_ORG"
|
||
}
|
||
}
|
||
|
||
RETURN_TYPES = (IO.IMAGE,)
|
||
FUNCTION = "api_call"
|
||
CATEGORY = "api node"
|
||
DESCRIPTION = cleandoc(__doc__ or "")
|
||
API_NODE = True
|
||
|
||
def api_call(self, prompt, seed=0, image=None, mask=None, n=1, size="1024x1024", auth_token=None):
|
||
model = "dall-e-2"
|
||
path = "/proxy/openai/images/generations"
|
||
request_class = OpenAIImageGenerationRequest
|
||
img_binary = None
|
||
|
||
if image is not None and mask is not None:
|
||
path = "/proxy/openai/images/edits"
|
||
request_class = OpenAIImageEditRequest
|
||
|
||
input_tensor = image.squeeze().cpu()
|
||
height, width, channels = input_tensor.shape
|
||
rgba_tensor = torch.ones(height, width, 4, device="cpu")
|
||
rgba_tensor[:, :, :channels] = input_tensor
|
||
|
||
if mask.shape[1:] != image.shape[1:-1]:
|
||
raise Exception("Mask and Image must be the same size")
|
||
rgba_tensor[:,:,3] = (1-mask.squeeze().cpu())
|
||
|
||
rgba_tensor = downscale_input(rgba_tensor.unsqueeze(0)).squeeze()
|
||
|
||
image_np = (rgba_tensor.numpy() * 255).astype(np.uint8)
|
||
img = Image.fromarray(image_np)
|
||
img_byte_arr = io.BytesIO()
|
||
img.save(img_byte_arr, format='PNG')
|
||
img_byte_arr.seek(0)
|
||
img_binary = img_byte_arr#.getvalue()
|
||
img_binary.name = "image.png"
|
||
elif image is not None or mask is not None:
|
||
raise Exception("Dall-E 2 image editing requires an image AND a mask")
|
||
|
||
# Build the operation
|
||
operation = SynchronousOperation(
|
||
endpoint=ApiEndpoint(
|
||
path=path,
|
||
method=HttpMethod.POST,
|
||
request_model=request_class,
|
||
response_model=OpenAIImageGenerationResponse
|
||
),
|
||
request=request_class(
|
||
model=model,
|
||
prompt=prompt,
|
||
n=n,
|
||
size=size,
|
||
seed=seed,
|
||
),
|
||
files={
|
||
"image": img_binary,
|
||
} if img_binary else None,
|
||
auth_token=auth_token
|
||
)
|
||
|
||
response = operation.execute()
|
||
|
||
img_tensor = validate_and_cast_response(response)
|
||
return (img_tensor,)
|
||
|
||
class OpenAIDalle3(ComfyNodeABC):
|
||
"""
|
||
Generates images synchronously via OpenAI's DALL·E 3 endpoint.
|
||
|
||
Uses the proxy at /proxy/openai/images/generations. Returned URLs are short‑lived,
|
||
so download or cache results if you need to keep them.
|
||
"""
|
||
def __init__(self):
|
||
pass
|
||
|
||
@classmethod
|
||
def INPUT_TYPES(cls) -> InputTypeDict:
|
||
return {
|
||
"required": {
|
||
"prompt": (IO.STRING, {
|
||
"multiline": True,
|
||
"default": "",
|
||
"tooltip": "Text prompt for DALL·E",
|
||
}),
|
||
},
|
||
"optional": {
|
||
"seed": (IO.INT, {
|
||
"default": 0,
|
||
"min": 0,
|
||
"max": 2**31-1,
|
||
"step": 1,
|
||
"display": "number",
|
||
"tooltip": "not implemented yet in backend",
|
||
}),
|
||
"quality" : (IO.COMBO, {
|
||
"options": ["standard","hd"],
|
||
"default": "standard",
|
||
"tooltip": "Image quality",
|
||
}),
|
||
"style": (IO.COMBO, {
|
||
"options": ["natural","vivid"],
|
||
"default": "natural",
|
||
"tooltip": "Vivid causes the model to lean towards generating hyper-real and dramatic images. Natural causes the model to produce more natural, less hyper-real looking images.",
|
||
}),
|
||
"size": (IO.COMBO, {
|
||
"options": ["1024x1024", "1024x1792", "1792x1024"],
|
||
"default": "1024x1024",
|
||
"tooltip": "Image size",
|
||
}),
|
||
},
|
||
"hidden": {
|
||
"auth_token": "AUTH_TOKEN_COMFY_ORG"
|
||
}
|
||
}
|
||
|
||
RETURN_TYPES = (IO.IMAGE,)
|
||
FUNCTION = "api_call"
|
||
CATEGORY = "api node"
|
||
DESCRIPTION = cleandoc(__doc__ or "")
|
||
API_NODE = True
|
||
|
||
def api_call(self, prompt, seed=0, style="natural", quality="standard", size="1024x1024", auth_token=None):
|
||
model = "dall-e-3"
|
||
|
||
# build the operation
|
||
operation = SynchronousOperation(
|
||
endpoint=ApiEndpoint(
|
||
path="/proxy/openai/images/generations",
|
||
method=HttpMethod.POST,
|
||
request_model=OpenAIImageGenerationRequest,
|
||
response_model=OpenAIImageGenerationResponse
|
||
),
|
||
request=OpenAIImageGenerationRequest(
|
||
model=model,
|
||
prompt=prompt,
|
||
quality=quality,
|
||
size=size,
|
||
style=style,
|
||
seed=seed,
|
||
),
|
||
auth_token=auth_token
|
||
)
|
||
|
||
response = operation.execute()
|
||
|
||
img_tensor = validate_and_cast_response(response)
|
||
return (img_tensor,)
|
||
|
||
class OpenAIGPTImage1(ComfyNodeABC):
|
||
"""
|
||
Generates images synchronously via OpenAI's GPT Image 1 endpoint.
|
||
|
||
Uses the proxy at /proxy/openai/images/generations. Returned URLs are short‑lived,
|
||
so download or cache results if you need to keep them.
|
||
"""
|
||
def __init__(self):
|
||
pass
|
||
|
||
@classmethod
|
||
def INPUT_TYPES(cls) -> InputTypeDict:
|
||
return {
|
||
"required": {
|
||
"prompt": (IO.STRING, {
|
||
"multiline": True,
|
||
"default": "",
|
||
"tooltip": "Text prompt for GPT Image 1",
|
||
}),
|
||
},
|
||
"optional": {
|
||
"seed": (IO.INT, {
|
||
"default": 0,
|
||
"min": 0,
|
||
"max": 2**31-1,
|
||
"step": 1,
|
||
"display": "number",
|
||
"tooltip": "not implemented yet in backend",
|
||
}),
|
||
"quality": (IO.COMBO, {
|
||
"options": ["low","medium","high"],
|
||
"default": "low",
|
||
"tooltip": "Image quality, affects cost and generation time.",
|
||
}),
|
||
"background": (IO.COMBO, {
|
||
"options": ["opaque","transparent"],
|
||
"default": "opaque",
|
||
"tooltip": "Return image with or without background",
|
||
}),
|
||
"size": (IO.COMBO, {
|
||
"options": ["auto", "1024x1024", "1024x1536", "1536x1024"],
|
||
"default": "auto",
|
||
"tooltip": "Image size",
|
||
}),
|
||
"n": (IO.INT, {
|
||
"default": 1,
|
||
"min": 1,
|
||
"max": 8,
|
||
"step": 1,
|
||
"display": "number",
|
||
"tooltip": "How many images to generate",
|
||
}),
|
||
"image": (IO.IMAGE, {
|
||
"default": None,
|
||
"tooltip": "Optional reference image for image editing.",
|
||
}),
|
||
"mask": (IO.MASK, {
|
||
"default": None,
|
||
"tooltip": "Optional mask for inpainting (white areas will be replaced)",
|
||
}),
|
||
},
|
||
"hidden": {
|
||
"auth_token": "AUTH_TOKEN_COMFY_ORG"
|
||
}
|
||
}
|
||
|
||
RETURN_TYPES = (IO.IMAGE,)
|
||
FUNCTION = "api_call"
|
||
CATEGORY = "api node"
|
||
DESCRIPTION = cleandoc(__doc__ or "")
|
||
API_NODE = True
|
||
|
||
def api_call(self, prompt, seed=0, quality="low", background="opaque", image=None, mask=None, n=1, size="1024x1024", auth_token=None):
|
||
model = "gpt-image-1"
|
||
path = "/proxy/openai/images/generations"
|
||
request_class = OpenAIImageGenerationRequest
|
||
img_binary = None
|
||
mask_binary = None
|
||
|
||
|
||
if image is not None:
|
||
path = "/proxy/openai/images/edits"
|
||
request_class = OpenAIImageEditRequest
|
||
|
||
scaled_image = downscale_input(image).squeeze()
|
||
|
||
image_np = (scaled_image.numpy() * 255).astype(np.uint8)
|
||
img = Image.fromarray(image_np)
|
||
img_byte_arr = io.BytesIO()
|
||
img.save(img_byte_arr, format='PNG')
|
||
img_byte_arr.seek(0)
|
||
img_binary = img_byte_arr#.getvalue()
|
||
img_binary.name = "image.png"
|
||
|
||
if mask is not None:
|
||
if image is None:
|
||
raise Exception("Cannot use a mask without an input image")
|
||
if mask.shape[1:] != image.shape[1:-1]:
|
||
raise Exception("Mask and Image must be the same size")
|
||
batch, height, width = mask.shape
|
||
rgba_mask = torch.zeros(height, width, 4, device="cpu")
|
||
rgba_mask[:,:,3] = (1-mask.squeeze().cpu())
|
||
|
||
scaled_mask = downscale_input(rgba_mask.unsqueeze(0)).squeeze()
|
||
|
||
mask_np = (scaled_mask.numpy() * 255).astype(np.uint8)
|
||
mask_img = Image.fromarray(mask_np)
|
||
mask_img_byte_arr = io.BytesIO()
|
||
mask_img.save(mask_img_byte_arr, format='PNG')
|
||
mask_img_byte_arr.seek(0)
|
||
mask_binary = mask_img_byte_arr#.getvalue()
|
||
mask_binary.name = "mask.png"
|
||
|
||
files = {}
|
||
if img_binary:
|
||
files["image"] = img_binary
|
||
if mask_binary:
|
||
files["mask"] = mask_binary
|
||
|
||
# Build the operation
|
||
operation = SynchronousOperation(
|
||
endpoint=ApiEndpoint(
|
||
path=path,
|
||
method=HttpMethod.POST,
|
||
request_model=request_class,
|
||
response_model=OpenAIImageGenerationResponse
|
||
),
|
||
request=request_class(
|
||
model=model,
|
||
prompt=prompt,
|
||
quality=quality,
|
||
background=background,
|
||
n=n,
|
||
seed=seed,
|
||
size=size,
|
||
),
|
||
files=files if files else None,
|
||
auth_token=auth_token
|
||
)
|
||
|
||
response = operation.execute()
|
||
|
||
img_tensor = validate_and_cast_response(response)
|
||
return (img_tensor,)
|
||
|
||
|
||
# A dictionary that contains all nodes you want to export with their names
|
||
# NOTE: names should be globally unique
|
||
NODE_CLASS_MAPPINGS = {
|
||
"OpenAIDalle2": OpenAIDalle2,
|
||
"OpenAIDalle3": OpenAIDalle3,
|
||
"OpenAIGPTImage1": OpenAIGPTImage1,
|
||
}
|
||
|
||
# A dictionary that contains the friendly/humanly readable titles for the nodes
|
||
NODE_DISPLAY_NAME_MAPPINGS = {
|
||
"OpenAIDalle2": "OpenAI DALL·E 2",
|
||
"OpenAIDalle3": "OpenAI DALL·E 3",
|
||
"OpenAIGPTImage1": "OpenAI GPT Image 1",
|
||
}
|